OpenAI has set 6 July 2026 as the day ChatGPT Workspace Agents leave their free preview and move to credit-based pricing. For an Australian small business that has been running pilots on the free tier, that date quietly turns a free experiment into a real line item. The sensible response is not to panic and not to rush a renewal. It is to put an honest number on what the agents will cost once the meter starts, and to set that figure next to the cost of building the same capability on Claude.
What changes on 6 July
Workspace Agents are the assistants that read your documents, act across your connected tools, and carry out multi-step tasks for a team. Through the preview period they have cost nothing to run, which made them easy to adopt and easy to overuse. From 6 July each agent run draws on credits. That single change moves your bill from a flat zero to a figure that rises and falls with how much work you push through the agents each month.
Every agent run consumes credits, so a busy month costs noticeably more than a quiet one.
Longer documents, more tool calls, and heavier tasks all increase the credits a single run uses.
Teams that relied on the free preview may see a bill appear where there was none before.
Usage is hard to forecast, which makes the new cost awkward for a small finance team to budget.
None of this makes the agents a poor choice. Plenty of Australian teams will get real value from them. It does mean the free-tier maths no longer applies, and any business about to commit a daily workflow to a metered service should know its likely annual cost first.
The real cost question for an Australian SMB
Credit pricing rewards light, occasional use and penalises heavy, daily use. A Sydney agency that runs a handful of agent tasks a week might spend $1,200 to $2,500 a year and barely feel it. A busier operations team running document processing through agents every day can reach $20,000 to $40,000 a year once seats and credit top-ups are added together. The distance between those two figures is the heart of the matter. Your cost depends on how hard your team pushes the tool, and that is precisely the part a vendor cannot quote you in advance.
This unpredictability is the practical problem. A flat zero needs no budget line. A figure that might be $1,500 one quarter and $12,000 the next forces a small business to either over-provision and waste money, or under-provision and risk a workflow stalling mid-month when the credits run low. For a lean Australian team without a dedicated finance function, that volatility carries a cost of its own, separate from the dollars on the invoice.
Light use, a few runs a week: often $1,200 to $2,500 a year, easy to absorb.
Steady daily use across a team: commonly $20,000 to $40,000 a year once credits are counted.
Spiky periods such as the end of financial year can lift the bill sharply with little warning.
Per-seat charges sit on top of credits, so growing headcount raises the baseline cost.
Where a Claude Cowork build compares
Claude Cowork lets a non-technical team hand real file and task work to Claude inside a desktop app, with the same agentic behaviour, connected tools, and multi-step handling that make Workspace Agents useful. The cost shape, though, is different. Rather than a per-run meter, a Cowork build is usually a fixed setup engagement followed by predictable ongoing access. For an Australian SMB that has quietly made agents part of its daily routine, a fixed shape is far easier to plan around than a meter that climbs every busy week.
A typical Automata AI Cowork setup is a fixed $3,500 engagement that maps your real workflows, connects your tools, and gets your team productive, after which the running cost stays steady rather than usage-linked. Set against a metered service that can reach $40,000 a year for a heavy team, that comparison is worth doing carefully before the free preview ends rather than after the first surprising invoice.
Predictable cost: a fixed build and a steady running cost, not a meter that moves with usage.
Data handling that respects the Privacy Act, with Australian compliance considered from the first conversation.
Claude as the engine, tuned to your actual processes instead of a generic template.
A named owner for the build, so the system is supported and improved rather than left to drift.
How to run the numbers before the deadline
The honest way to decide is to measure your own usage rather than trust a headline price. Estimate how many agent runs your team really does in a normal week, multiply that across the year, and set the metered cost beside a fixed-build alternative. For a genuinely light user, staying on a credit plan may well be the cheaper and simpler answer, and we will say so. For a team that has made agents part of its daily work, a Claude Cowork build often lands lower over twelve months and takes the guesswork out of the budget line.
If you want help putting a real figure on your team's agent usage before 6 July, book a brainstorm and we will run the comparison on your numbers rather than a generic estimate.
